# Confirmed User Memory

> Facts about Tony that are confirmed and durable.

| Fact | Source | Date |
|------|--------|------|
| Runs a local FreeToken LLM server (RTX 3060 12GB) at `:1919/v1` with gpt-oss-20b and Gemma-4-26B. Manual-start only due to VRAM conflict with LFM2.5. Wired into Hermes as `custom:freetoken`. | Conversation | 2026-08-30 |
| Prefers fast cloud models over slow local 26B for CLI/voice to avoid ~1-minute reply delays. | MEMORY tool | 2026-08-30 |